Petersen Graph Multi-Orientation Based Multi-Scale Ternary Pattern (PGMO-MSTP): An Efficient Descriptor for Texture...

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Texture classification is challenging due to variations in rotation, illumination, scale, and viewpoint.
• Existing methods like Local Graph Structure (LGS) and Local Ternary Patterns (LTP) have limitations.

Purpose of the Study:

• To propose a novel image feature descriptor, Petersen Graph Multi-Orientation based Multi-Scale Ternary Pattern (PGMO-MSTP), for robust texture and material classification.
• To overcome the shortcomings of existing LTP-like and LGS-like descriptors.

Main Methods:

• Developed single-scale Petersen Graph-based Ternary Pattern descriptors (PGTPh and PGTPv) encoding 5x5 image patches.
• Utilized Petersen graph-shaped sampling structures to capture pixel relationships.
• Combined histograms from PGTPh and PGTPv to create the multi-scale PGMO-MSTP model.

Main Results:

• PGMO-MSTP demonstrated superior performance against state-of-the-art handcrafted and deep learning-based texture descriptors.
• Experiments on sixteen challenging texture datasets confirmed the effectiveness of the proposed method.
• Statistical analysis using the Wilcoxon signed rank test indicated PGMO-MSTP as the top-performing descriptor across all datasets.

Conclusions:

• The proposed PGMO-MSTP descriptor offers a robust and effective solution for texture and material classification, especially under challenging image variations.
• PGMO-MSTP represents a significant advancement in texture analysis, outperforming current leading methods.